y Traf Þ c Information Dissemination in Vehicular Ad Hoc ... · Traf c and Travel Information (TTI) spreading in vehicular ad hoc networks is achieved by the means of a oodingmechanism

0

yTraffic Information Dissemination in Vehicular AdHoc Networks

Attila Torok, Balazs Mezny and Peter LaborcziBay Zoltan Foundation For Applied Research,

Institute for Applied Telecommunication TechnologiesHungary

1. Introduction

Today, cars are equipped with all kind of on-board sensors and microcomputers that are ableto measure geolocation, speed, tire pressure, raindrops on the windshield, etc., and basedon these information Intelligent Transportation Systems (ITS) are built. The ITS applicationsare intended to ease the everyday life of drivers by reducing the risk of accidents, improvingsafety, increasing road capacity and reducing traffic jams. Many research papers, for exampleTorok et al. (2008) and Sormani et al. (2006), pointed out that a significant reduction of trafficjams can be achieved through the use of vehicular ad-hoc networks (VANETs). Vehiclescould serve as Traffic and Travel Information (TTI) collectors and transmit this information toother participants in the vehicular network Laborczi et al. (2006). The ITS applications couldutilize this information to actively relieve traffic congestion. Practically, vehicles could detecttraffic congestion automatically when the usual (historical) characteristics of traffic patternsdrastically change, i.e. the number of neighboring vehicles is high and/or the average speedis too low. Then this information should be relayed, disseminated to the vehicles approachingthe congested area; thus, they will have enough time to choose alternative routes.Due to their inherent characteristics, viable communication is harder to support in ITSscenarios than in conventional wireless networks. Vehicles are usually moving much fasterthan traditional mobile nodes; moreover, a vehicular network might be very heterogeneous interms of node density, becoming fragmented in many cases. Reliability is also compromiseddue to the usually high interference in urban scenarios. Thus, there is a need to reconsider thewireless ad hoc communication networking protocols, and to use new concepts that fit betterthe specificities of ITS applications.Traffic and Travel Information (TTI) spreading in vehicular ad hoc networks is achieved bythe means of a flooding mechanism. To overcome network fragmentation the vehicles usuallymaintain and carry a copy of the packets, which is disseminated along the road segmentsZhao & Cao (2006), Burgess et al. (2006), Tian et al. (2004). The frequency of subsequenttransmissions will control the quality of the TTI reports, in terms of delay and accuracy. Ifthe frequency of TTI transmissions is high, the time necessary for the information to reach theouter bounds of the geographic area is lower. The accuracy of TTI also varies in function ofthe amount of communication involved in the travel information gathering and transmission.Frequent information exchange leads to a more accurate picture about the traffic situation,but also to superfluous dissemination. Superfluous forwarding can be reduced by usingadaptivity in the flooding mechanisms. Adaptivity can be introduced by controlling the

6

www.intechopen.com

2 Theory and Applications of Ad Hoc Networks

frequency of information exchange (timelymanner) or limiting the dissemination only to areaswhere the TTI is really necessary (spatial manner).Besides the presentation of the most important spatial TTI dissemination protocols we alsoinvestigate the problem of determining the areas of interest of traffic jams. As we argue, thepresented spatial dissemination protocols fail to properly define the places where the TTI isuseful. These solutions are only effective when are employed with additional mechanisms,which provide context-aware information to calculate the areas of interest of specific trafficjams.

2. Literature review

This section presents protocols related to spatial adaptivity-based TTI dissemination,which can be achieved pro-actively, using a data-push model Sormani et al. (2006),Leontiadis & Mascolo (2007), or based on a data-pull model Dikaiakos et al. (2007), when theinformation is obtained on-demand. In the first case the data is usually disseminated from thetraffic incidents towards the outer inbound road segments, while in the second case the data ispulled to the locations of interest on-demand. In both cases the question is how to control andlimit the traffic information dissemination only to places where the respective information isuseful.

2.1 Spatial adaptivity by using data-push protocols2.1.1 Dissemination restricted through publish/subscribeThe possibility of restricting the TTI dissemination to certain areas is investigated inLeontiadis & Mascolo (2007). In their proposal the authors present a publish-subscribemethod, as the members of the traffic will receive only messages of their interest. The solutionworks well with methods employing the data-push model, for example the one described inSormani et al. (2006). The publish-subscribe process starts with a vehicle subscribing to a topic(e.g. traffic congestion information). When a vehicle publishes a message, the area of interestand validity time of the message is determined. Vehicles subscribed to the given topic willreceive the message if they are within the area of interest and the message is still valid. Thebasic idea is to maintain the message in the notification area, so every vehicle approachingthe area where the message was generated (for example a traffic accident) gets the notificationand has a chance to consider its reaction to the event (e.g. taking an alternate route to itsdestination). This is achieved by generating a fixed number of replicas of the message, whichmeans that only those vehicles will broadcast themessagewhich have a replica of themessage.This way the load of the communication network is reduced compared to the general floodingmechanism, where every node of the network retransmits the receivedmessage, resulting in abroadcast storm. If a vehicle carrying a replica of the message is leaving the notification area,then it hands over the replica to an other vehicle, preferably driving the opposite direction, tokeep the message replica in the desired area.There are two questions regarding the message replicas. How many replicas should be there,and who should carry them? Before the replica owner broadcasts the message, it poll itsneighbouring vehicles regarding the topic of the message. There are three possible answers tothis poll:

– Informed: The answering vehicle is already received a notification for the given topic (e.g.if the topic is parking spots, this vehicle already knows where are free parking spaces).

– Interested: This vehicle is subscribed to the topic.

108 Mobile Ad-Hoc Networks: Applications

www.intechopen.com

yTraffic Information Dissemination in Vehicular Ad Hoc Networks 3

– Not interested: The vehicle is not subscribed.

If there are interested vehicles the carrier broadcasts the message. Also if the carrier is leavingthe designated area, it selects a new carrier heading for the notification area, with the mostinterested vehicles in its vicinity. The aim of this selection method is to get the messagereplicawhere themost uninformed vehicles come from. This mechanism results in the replicasconverging to areas where the information is needed, and if there are two replicas in the samearea, one of them will move to an other area where the message is needed.The number of the replicas is determined adaptively. Every replica carrier keeps the result ofthe last k polls, and based on these statistics the following options are possible:

– If there was at least one uninformed subscriber in the last k polls, the replica is kept.

– If there were at least k uninformed subscribers then a new replica is generated andforwarded to a vehicle, determined by the new carrier selection mechanism.

– If there are no uninformed subscribers, the replica is marked for deletion. In order to avoiddeleting replicas simultaneously, the replicas are merged and are deleted only if the carrierreceives a broadcast from an other carrier.

This way the number of replicas are adapted to the demand for the message, and they areforwarded to areas with the most subscribers.However, due to the carrier selection and TTI replication mechanisms, it is not alwaysguaranteed that the information carriers will meet their subscribers. The chance that a replicasurvives insensitivity, and meets proper subscribers, depends on the estimate of the replica’snecessity, which is represented by the number of last k polls. Thus, the successful outcome ofthe protocol highly depends on the topological context and the fine tuning of the system. Forexample, considering the simulation results presented in Figure 1 for a scenario where twointersections are interconnected through two one way roads (one with traffic jam), it turns outthat the fraction of cars entering the jammed road depends highly on the frequency of TTIdisseminations (Timer), respectively on the number of transmissions until a TTI remains alive(TTL). If the frequency is too high then the TTImessage is not transported until the intersectionwhere the vehicles must be informed, even considering higher values for TTL. This can beattributed to the fact that the TTI replication and propagation was determined based on theinterest of other neighboring vehicles, and in this particular case all the vehicles are headingoutwards the jammed area; thus, they are uninterested about this particular jam. In order toovercome such problems additional context information regarding the road infrastructure hasto be taken in consideration.

2.1.2 Dissemination restricted through propagation functionsIn Sormani et al. (2006) the authors investigate methods on how to propagate the trafficmessages to areas where the respective information is useful. They outline a scenario, wherean accident occurs on a highway. A message broadcast happens within one mile of theaccident, telling the vehicles to slow down. A second message is delivered to key pointsof the highway, where drivers can take alternative routes, in this scenario these points arethe highway exits. This method can be considered as a data-push model, where the messageis disseminated even if the information wasn’t requested by an entity. The main idea is thedefinition of a propagation function, which restricts the message propagation to areas wherethe message is important. For our example this represents the highway, there is no point indisseminating the message outside this area. This propagation function has minimum pointsat the target zones, and its value is increasing as the distance from the target zones increase.

109Traffic Information Dissemination in Vehicular Ad Hoc Networks

www.intechopen.com


Fig. 1. Effect of TTI replication on alternative route selection

The message, originated the place of an event (e.g. a traffic accident), is always forwarded toa vehicle whose position results in the lowest value in the propagation function. This way themessage will be routed towards the minimum of the propagation function, the target zone.The shape of this function is determined to follow the road network, where the vehicles candisseminate themessage. The propagation function is either computed by the originator of themessage, taking into account the current traffic situation and the road network in the vicinity,or it can be precomputed for important areas.The authors consider some basic protocols to disseminate the traffic message in order toevaluate the effects of the propagation function. The most basic protocol is a modification ofthe flooding mechanism, where the received message is rebroadcasted only for the first timeit has been seen and the value of the propagation function at the receiving vehicle is lowerthan at the sender of the message (One Zero Flooding, OZF). An other basic protocol is afurther modification, taking into account the distance between the sender and the receiver(Distance Driven Probabilistic Diffusion, DDPD). This distance is used for probabilisticmessage forwarding, where the probability of forwarding is the distance between the vehiclesdivided by the communication range (approximately 200 meters for 801.11 capable devices).This way the surplus message retransmissions can be avoided. A more advanced protocoltakes into account the shape of the propagation function (Function Driven ProbabilisticDiffusion, FDPD). In this case the probability of forwarding is zero at the sender’s positionand is increasing as the value of propagation function decreases, and takes the value ofone at the lowest value of the propagation function inside the communication radius of thesender of the message. This method yields to a more accurate routing, as a lower valueof the propagation function is not enough, the algorithm tries to find the lowest possiblevalue. The authors propose some store & forward variations of these algorithms, whereafter receiving a message the vehicle not only retransmits it immediately but carries it forsome time and rebroadcasts the message periodically. The first store & forward variant(Function Driven Feedback-augmented Store & Forward Diffusion, FD-FSFD) is based onFDPD with the addition of a feedback augmented store & forward mechanism. The feedbackaugmention means, if the carrier receives a message from a vehicle whose position resultsin a lower value of the propagation function, then the further broadcasts are cancelled asthe message already reached a lower point of the propagation function. The second store &forward algorithm (Direction-aware Function Driven Feedback-augmented Store & Forward


www.intechopen.com


Diffusion, DFD-FSFD) is an extension of FD-FSFD by taking into account the direction ofmovement of the nodes. This means that only nodes moving towards lower points of thepropagation function are used to carry the message. These methods are useful in sparsenetworks where the connection between clusters of vehicles is not guaranteed.Unfortunately, there are no methods presented to calculate the propagation function, i.e., thelocations where the information should be propagated. Therefore, this protocol is not readyto be applied for TTI dissemination in urban scenarios.

2.2 Spatial adaptivity by using data-pull protocolsIn Dikaiakos et al. (2007) the authors outline an application-layer communication protocol(Vehicular Information Transport Protocol, VITP), which could be used in VANETs todisseminate location based information. Such location based information can be trafficinformation regarding road conditions (e.g. slippery road or congestion), or some kind ofroadside service information (e.g. fuel prices at gas stations or menus of restaurants). Thesekinds of information are typically requested by someone; thus this method can be called as thedata-pull model. The authors introduce the concept of virtual ad-hoc servers (VAHS), whichmeans that an information request is processed by a number of peers at the target location ofthe request, and the result is sent back to the originator of the query. For example, if a vehiclewants to know the traffic condition on a road segment in its path, it sends a request to thatroad segment. When a vehicle in the target area receives the query, it attaches the requestedinformation to the message, and retransmits the message, so other vehicles can contributeto the reply. The ones contributing to the reply constitute the virtual ad-hoc server. After acertain threshold is met, for example ten vehicles attached their velocity information to themessage, the last vehicle generates the reply from the gathered data, and sends it back to theoriginator vehicle. This way the answer can be more accurate, than in the case where only onevehicle made the reply, or when separate replies were generated by multiple vehicles. Thedata-push method is also supported by the proposed protocol as it is favorable in some cases,for example in case of an accident. The vehicles couldn’t be forced to constantly generatequeries for accidents, instead the information is “pushed” to them. The described protocol isalso capable of caching the information in some cases, so a reply could be made before thequery reaches the target location, speeding up the process. The effect of caching is furtherelaborated in Dikaiakos et al. (2010), and it is shown that significant improvements can beachieved in both the data-pull and the data-push cases.

2.3 Aggregation scheme for roadside unit placementThe authors of Lochert et al. (2008) present a method for optimization of roadside unitplacement in order to minimize the required bandwidth for traffic information dissemination.A domain specific aggregation scheme is presented, then a genetic algorithm is proposedto identify the most appropriate positions for the roadside units. The aggregation schemeis conceived in a hierarchical fashion: the farther away a region is, the coarser will bethe information on its traffic situation. By using this scheme a vehicle traveling along theroad network will obtain coarser and coarser approximations about the traffic situation,travel times will be summarized in regions that are farther and farther away. Thus, theaggregation scheme will allow to limit the bandwidth requirements for TTI dissemination, byreducing the network capacity necessary for information spreading. The aggregation schemeis based on the definition of special multi-level landmarks, which will cover the hierarchyof the road networks. The higher levels are constituted by highways or junctions of main


www.intechopen.com


roads, while the lower levels will include all higher level landmarks plus more and moreintersections of smaller streets. Thus, cars can make investigations about the travel timesbetween neighboring landmarks, which informationwill be disseminated in the surroundingsof the respective road segment. A coarser picture, calculated from travel times betweenlandmarks of higher levels, will be disseminated to a larger area, which leads to a summarizedview of the travel times in the area. Roadside units are placed to form a backbone network,allowing them to exchange the TTI to be disseminated. In order to use a very limitednumber of roadside units the authors propose a toolchain for placement optimization. Sincethe identification of the optimal subset of roadside unit locations is a difficult optimizationproblem a genetic algorithm based approximation method is used to obtain a good resultsubset. The toolchain is composed from a network and traffic simulator (ns-2 and VISSIM),respectively from a closely interacting application simulator and the genetic algorithm. Theapplication simulator is used to process the log file of the network-traffic simulator, performthe specific aggregation methods, decide about the dissemination of TTI beacons. At the levelof the network simulator all the possible roadside unit locations are simulated, all of themtransmit the periodic beacons. The non-existing roadside units are ignored at the level ofthe application simulator, the received beacons are not considered when its knowledge baseis updated by the genetic algorithm. Thus, with the separation of movement and networkissues from application behavior travel time savings are achieved by calculating the vectorsof active roadside unit locations. These savings are used as a fitness metric, making theapplication-centric optimization through the genetic algorithm. The viability of the approachis confirmed through simulations by applying the proposed solution to a large-scale cityscenario.

3. Spatially-aware congestion elimination (SPACE)

In this section the SPatially-Aware Congestion Elimination algorithm (SPACE) is designed.An algorithm is given to determine the locations, domain of interests, where a possible event(e.g. traffic jam) on a certain road influences the route choice of the driver. To illustratethe problem, a small example is presented, then we formulate it as a graph theoreticaloptimization problem. Both a heuristic and a linear programming optimization solution areprovided. Thus, we give a well defined area (the Domain of Interest, DoI) where informationabout a specific traffic jam is useful.

3.1 ExampleFirst let us consider an example of one way roads from left to right (Figure 2), which representsa subset of a larger road network.We assume that a vehicle enters the network at node 1, its destination is at node 10. The vehiclehas route decisions at nodes 2,3,4 and 5, respectively. It can take either Route A, Route B, RouteC or Route D to reach its destination. Route A is shortest and fastest; consequently, the vehicletakes the middle route in the default case. If route A at road segment 6-7 is congested, thisinformation has to be disseminated throughout the road network.TheDomain of Interest (DoI) is defined as the set of road segments, where the information abouta traffic jam influences the route choice of the driver, i.e., the roads where the informationshould be disseminated. At these places, the vehicles are still able to change their routes,without a drastic deterioration in their travel time. However, if the vehicle leaves a criticaljunction, enters in the zone of no return, where is no possibility to avoid the traffic jam, or onlywith a major increase in the travel time. Our scope is to optimize the area of DoI in order to


www.intechopen.com


Fig. 2. Example road network

reduce the amount of TTI flooding and at the same time to achieve as low vehicle travel timesas possible.Traditional flooding methods disseminate this information towards any directions. However,in the example this information is only interesting at the decision point 4 (optimized DoI),since the second best choice is route B. It is useless to deliver the TTI further than junction4, as vehicles are heading towards junction 4, anyway. There is no sense in providingthis information to the whole DoI, like (1,2,3,8,9). However, if both routes A and B wouldbe congested, this information should be provided to an earlier decision point (junction 2,segment 1-2), where both routes can be avoided by the by-pass route C. This means that theDoI can also present characteristics varying over the time.

3.2 Problem formulationThe road network is represented by a directed and weighted graph G(V ,E) withrepresentation described in Speicys & Jensen (2008), using two types of edges Er and Et(Er ∪ Et = E , Er ∩ Et = ∅). Er is a directed edge representing a road between two intersections.One-way roads are represented with directed edges, while two-way roads with two oppositedirected edges. The set of Et represents the turning regulations, i.e., an edge from n1 ∈ V ton2 ∈ V , where n1 is the destination node of e1 while n2 is the origin node of e2, is included inthe graph if and only if a turn is allowed from e1 to e2. The weight of an edge represents thetravel time on the corresponding road, or turning.The event (traffic jam) is associated to a set of failed roads E f , which is a subset of the roads(E f ⊆ E ). We assume that the set E f contains the core of the problem, where the actual speeddecreases to a fraction of the normal speed.We also assume that an estimated Origin-Destination (OD) function for the road network isknown. The OD function OD(n,m) represents the average amount of vehicles traveling fromnode n ∈ V to m ∈ V . If the OD function is not known then it can be assumed that it hasuniform distribution, i.e., OD(n,m) = 1 for each n,m ∈ V .The output of the algorithm is an Impact Vector IE f

(e) that shows whether an event on edges

of E f has an impact on the route choice of vehicles travelling on edge e, and if yes, in whatextent. The value of IE f

(e) is zero if it has no impact on edge e, non-zero if it has an impact.


www.intechopen.com


The value IE f(e) expresses the average amount of vehicles on edge e, whose route choice is

impacted by the knowledge about an event on roads of E f .

3.3 SPACE algorithmIn this section the proposed SPACEAlgorithm is described, which generates the Impact VectorI. The algorithm simulates the impact of an event (obstacle, traffic jam) on a set of edgesof the graph. For each affected optimal path p of the graph it is assumed that the traveltime on edges of E f increases significantly. Then, the weights of the affected edges of E f

are increased, or these edges are excluded from the graph temporarily, and a new optimalpath (the by-pass route) is calculated by running the shortest path algorithm in the temporarygraph, resulting in path p∗. We define three parts of the optimal route p, and illustrate iton Figure 2, where p=(1,2,3,4,5,6,7,10), and the alternative route bypassing the edges withincreased weights p∗=(1,2,3,4,8,7,10):

Part I. Set of edges common with p∗, before the disjoint part: e.g., road (1,2,3),

Part II. Set of edges not included in p∗, before the event: e.g., road (4,5,6),

Part III. Set of edges after the event: e.g., road (7,10).

In this case, an event on edges in E f is important for the last X of edges in part I of p (inorder to choose another route), and in the disjoint part before the event (part II.), in order tobe informed about the obstacle (without the possibility of choosing the other route). Theseedges are called relay edges in the algorithm. For all these edges the impact vector I has to beincreased with the amount of vehicles traveling on that route (or by 1, if an OD matrix is notavailable). The algorithm is summarized as follows:

Input: Directed Weighted Graph G ,OD matrix, Set of failed roads E f

Output: Impact Vector Ifor all Pair of nodes (n,m) ∈OD do

Calculate in G the optimal path p from n to m;Create a new temporary graph GE f

: increase weights of edges of E f significantly;

Calculate by-pass route p∗ from n to m in GE f;

Calculate the set of relay edges: ER ⊆ E ;for all edge er ∈ ER do

Increase IE f(e) by OD(n,m);

end forend for

3.4 SPACE ILP algorithmIn this subsection the problem of finding the optimal Domain of Interest (DoI) is formulated asan Integer Linear Programming (ILP) problem, as presented in Torok et al. (2010). Although,solving an ILP by a solver has a long running time, we emphasize that this formulation hasthe following motivations: the formulation gives an exact definition of the TTI disseminationproblem and it allows a precise analysis of the problem compared to heuristic algorithms.First, let us define the normal route of the vehicles. For each edge (i, j), i, j ∈ V , and originand destination nodes n,m ∈ V , we define the set of assignment variables, X = {xnmij }. The

variable xnmij takes value 1 if edge ij is used in the shortest path from n to m, and 0 otherwise.

The known flow conservation constraints for the default routes are as follows:For each j,n,m ∈ V whereOD(n,m)> 0:


www.intechopen.com


∑i∈V

xnmij − ∑k∈V

xnmjk =

{

−1 if j = n1 if j = m0 otherwise

(1)

Similarly, the by-pass route is defined for the vehicle. For each edge (i, j), i, j ∈ V , and originand destination nodes n,m ∈ V , we define the set of assignment variables, Y = {ynmij }. The

variable yij takes value 1 if edge ij is used in the by-pass route from n to m, otherwise 0. Theflow conservation constraints for the by-pass routes are as follows:For each j,n,m ∈ V whereOD(n,m)> 0:

∑i∈V

ynmij − ∑k∈V

ynmjk =

{

−1 if j = n1 if j = m0 otherwise

(2)

Furthermore, in the formulation, both the normal and the by-pass routes are to be split inseveral pieces. For this, five more assignment variables are defined: anmij , bnmij , cnmij , dnmij ,

f nmij (for each edge (i, j), i, j ∈ V and origin and destination nodes n,m ∈ V) with following

definitions:

– anmij is 1 if edge ij belongs to the common part of the normal and the by-pass route, 0

otherwise.

– bnmij is 1 if edge ij belongs to the normal route after the fork of the normal route but before

the traffic jam, 0 otherwise.

– cnmij is 1 if edge ij belongs to the traffic jam ((i, j) ∈ E f ), 0 otherwise.

– dnmij is 1 if edge ij belongs to the normal route after the fork of the normal route but after the

traffic jam, 0 otherwise.

– f nmij is 1 if edge ij belongs to the by-pass route while not to the normal route, 0 otherwise.

For an example of these definitions see Figure 2 with origin=1, destination=10, optimal route(1,2,3,4,5,6,7,10) and by-pass route (1,2,3,4,8,7,10). anmij = 1 for roads (1,2),(2,3),(3,4) and (7,10).

bnmij = 1 for roads (4,5) and (5,6). cnmij = 1 for road (6,7). dnmij = 0 for all roads. f nmij = 1 for roads

(4,8) and (8,7).The above definitions are ensured by the following equations:For each (i, j) ∈ E and n,m ∈ V whereOD(n,m) > 0:

anmij + bnmij + cnmij + dnmij = xnmij (3)

xnmij + f nmij ≤ 1 (4)

anmij + f nmij = ynmij (5)

Furthermore, the part of the default route after the jammed link (d) has to be distinguishedform the part before the jam (b) with the following constraint:For each j,n,m ∈ V whereOD(n,m)> 0:

∑i∈V

dnmij − ∑k∈V

dnmjk =

{

−1 if cnmij = 1

≥ 0 otherwise(6)


www.intechopen.com


For each road (i, j) affected by the traffic jam ((i, j) ∈ E f ) set cnmij to 1 and ynmij to 0, while for

each other (not jammed) road ((i, j) /∈ E f ) set cnmij to 0.

Next, the assignment variables for the propagation region are defined and we formulate thefact that vehicles does not by-pass the jam until they receive amessage about it, i.e., the normalroute and corresponding by-pass route are to be the same outside the propagation region. Foreach edge (pair of nodes) (i, j), i, j ∈ V , we define the set of assignment variables, R = {rij}.The variable rij takes value 1 if edge ij is included in the propagation region, otherwise 0.In order to ensure a propagation region that reaches all places where normal and by-passroutes are to be forked, the following constraints are defined:For each n,m ∈ V where OD(n,m)> 0:

rij ≥ bnmij (7)

Finally, we define the objective by minimizing the weighted average of the length of allby-pass routes and the total length of the propagation region:

min ∑(i,j)∈E

(

αl ′ij ∑n,m∈V

ynmij + (1− α)l ′′ijrij

)

(8)

l ′ij denotes the cost (length, travel time, etc.) of travelling on road ij while l ′′ij denotes the

cost (e.g., road length, communication cost) of propagating information on road ij. Parameterα (0 ≤ α ≤ 1) expresses the importance of minimizing the total length of all by-pass routesagainst the total propagation region.In summary, for the ILP formulation we define constants: cnmij , l ′ij, l

′′ij ; binary variables xnmij ,

ynmij , rij, anmij , bnmij , dnmij , f nmij ; objective: (8) and constraints: (1)-(7).

3.5 Query-based information gatheringAs we presented above, in Dikaiakos et al. (2007) a query-based protocol is provided toachieve spatial adaptivity by the means of pull-based techniques. Unfortunately, there isno exact mechanism specified to calculate at what extent the queries, respectively the TTIreply/caching, should be propagated inside the road network. In the original paper theauthors present simulations, where the queries are propagated only to a randomly selectedvalue of 400-800 meters inside the road segments. This means that in certain situations theinformation will not be received in time to calculate the proper by-pass route. Therefore,additional mechanisms are necessary to determine the critical points until when the queriesmust be propagated, i.e., from where the TTI information has to be gathered. Considering avehicle entering at junction 1 in the road graph of Figure 2, it is hard to decide when and howdeep to inject the traffic information query in order to discover a possible traffic jam on routeA. This problem is related to the calculation of the optimal DoI, and can only be solved byusing additional information about the vehicle’s context, i.e. the route graph traversed duringthe trip.In Laborczi et al. (2010) an extension of the previous ILP formulation is given, with the aim tooptimize the positions where vehicles along their routes should send querymessages, in orderto collect information about possible traffic jams. Thus, this formulation can be considered asan optimization of the query protocol presented in Dikaiakos et al. (2007). The numericalresults of the formulation are presented in the following section.The equations 1 - 6 from the previous section are the same also for the query-based informationgathering.


www.intechopen.com


In order to find an optimal point to send a query message, we have to define the following setof variables:

– snmij is 1 if edge ij belongs to the common part of the normal and the by-pass route and to

the DoI as well.

With the following three constraints the properties of variables snmij are ensured:

For each n,m ∈ V where OD(n,m)> 0:

rij ≥ bnmij + snmij (9)

snmij ≤ anmij (10)

Edges where snmij or bnmij takes value 1 should be a coherent region, which ends at the traffic

jam:

∑i∈V

(snmij + bnmij )− ∑k∈V

(snmjk + bnmjk ) =

{

1 if ∃k cnmjk = 1≤ 0 otherwise

Finally, the propagation delay of the distributed messages has to be taken into account. Ittakes some time for the query message to reach the jam, and the reply message to reach theoriginator. The querymessage should reach the begin of jam, collect information and the replymessage should reach the vehicle before it reaches the optimal decision point expressed by thefollowing constraint:

∑(i,j)∈V

(snmijlij

vveh− (2bnmij + snmij )

lij

vmess)≥ 0, (11)

Where lij is the length of the ij road segment, vveh is the velocity of the vehicle, vmess is thevelocity of the message propagation.Finally, we define the objective by minimizing the weighted average of the length of allby-pass routes and the total length of the DoI:

min ∑(i,j)∈E

(

αl ′ij ∑n,m∈V

ynmij + (1− α)l ′′ijrij

)

(12)

where, l ′ij is the cost (length, travel time, etc.) of traveling on road (i, j)while l ′′ij denotes the cost

(e.g., road length, communication cost) of propagating information on road (i, j). Parameterα (0 ≤ α ≤ 1) expresses the importance of minimizing the total length of all by-pass routesagainst the total DoI.In the next section, we use this formulation as follows. First, for each n,m ∈ V whereOD(n,m) > 0 calculate shortest path and set variables xnmij based on the result of the shortest

path algorithm. Second, solve the following ILP problem: constants: xnmij , cnmij , l ′ij, l′′ij; binary

variables: ynmij , rij, anmij , bnmij , dnmij , f nmij , objective: (12) and constraints: (2)-(6) and (9)-(11).

4. Numerical analysis of the SPACE and SPACE ILP algorithms

In this section we present the evaluation of the proposed heuristic and linear programmingalgorithms. All the simulations were effectuated on the same section of a digital map ofBudapest.


www.intechopen.com


4.1 SPACEThe output of the SPACE algorithm is demonstrated on Figure 3 for two roads of the Budapesttest network. A main road (bridge, solid line), and a side road (from down-town, dotted line)were considered for analysis. The x-axis represents the domain of interest, i.e., the sum ofroad lengths on which the information about the event is disseminated, while y-axis representthe impact factor. The information is sent to roads (e) of higher impact factors (IE f

(e)), while

roads with minor impact factors can be neglected. First, we assume that there is a traffic jam(obstacle) on the main road (depicted with solid line). If the TTI information is flooded to thewhole domain of interest, then it should be spread to 16,000 meters. Therefore, a threshold(e.g., TR = 4,000) should be set in order to avoid superfluous forwarding. In this way theinformation is carried to the majority of vehicles (just 1,000 from more than 15,000 do notreceive the information in time), while the domain of interest is decreased to 4,000 meters,instead of 16,000. Second, we consider an obstacle on a side street (dashed line). As expected,the impact factor of such streets is less, i.e., if the threshold is set to 1,000 then the domain ofinterest is 500 meters.

Fig. 3. Domains of Interest for different road types using the SPACE algorithm

4.2 SPACE ILPIn order to have a better understanding, the results of the SPACE ILP algorithm are presentedbelow. A main road ( bridge), and different side roads (from downtown area) were consideredfor analysis. The bridge graph represents averaged values for traffic demands initiated fromboth sides of the city, considering traffic jam on one of the bridge lanes. The downtown graphrepresents averaged values from different congested downtown roads (considering also themajor roads leading to the bridge).Figure 4 shows the Domain of Interest (DoI) depending on the parameter α (see Objective(8)). We recall that α (0 ≤ α ≤ 1) expresses the importance of minimizing the total length ofall vehicle by-pass routes against the importance of minimizing the propagation region (areaof dissemination). The DoI is represented as the sum of the road segment lengths included inthe propagation region.It is obvious that for both graphs the DoI increases by increasing α. On the other hand, thefigure shows that the two types of roads represent different dynamics considering their DoI.In case when the obstacle is on the bridge, the DoI increases steeplywith the increase of α. Thismeans that in order to reach all roads of the maximumDoI, higher effortsmust be involved forTTI dissemination. However, after a limit (α ≥ 0.6) the DoI is not increasing significantly (onlyabout 1 km). A crucial point for α is between 0.3 - 0.4, where the DoI increases significantly.Considering congestion on downtown roads the situation is different. It can be seen, that the


www.intechopen.com


Fig. 4. Domains of interest in function of parameter α

variance of the DoI values is higher; however, the area of DoI for downtown scenarios is onlya fraction of the values of the bridge scenarios.These observations are also validated if we consider the length of alternative (by-pass) routesin function of α (Figure 5). As α increases, the length of by-pass routes will decrease, becausemore and more vehicles will be able to choose the ideal by-pass routes to avoid the congestion.For the bridge scenario the length of alternative routes decreases with about 30% if wedisseminate TTI by employing α = 0.4. In case of downtown congestions, we can observethat the length of alternative routes will not decrease significantly as we increase α, since thebest by-pass routes are closer to the area of congestion. Thus, for downtown roads it is uselessto disseminate the information further than the next couple of road segments (e.g. 200-300meters), since the by-pass routes would not become shorter in any case.Numerous analysis have been carried out that also show that the effect of α on the DoI andlength of alternative routes is significant between 0.2 and 0.4 for most of the roads.

4.3 Query-based information gatheringIn this section we present the numerical results of the generic query-based traffic informationgathering protocol. The results were generated by creating a large amount of randomsource-destination route pairs on the road graph of Budapest. Optimal query locations weregenerated by solving the ILP formulation described in the previous section. A characteristicset of results, presenting interesting cases, were selected for presentation.On Figure 6 the x-axis represents the distance of the source (n) of a vehicle from the traffic jam(while the destination (m) was fixed), while the y-axis represents the alteration in length ofthe respective metrics. The road length increase presents the difference between the originaland the different by-pass routes (∑(i,j)∈E l

′ij(y

nmij − xnmij ), for source n and destinationm), while

the query distance metric presents the distance from where the query is injected towards thepoint of interest (∑(i,j)∈E(l

′′ijrij), l

′ij = l ′′ij = length of road (i, j)).

In case when the jam was on a bridge (diagrams noted with (B)), the length of the originalroute was 3300 meters. When the distance was less than 200 meters from the traffic incident,the increase in the by-pass route length was nearly 1600 meters, an increase of around 50% of


www.intechopen.com


Fig. 5. Length of alternative routes in function of parameter α

the original route. As the source was generated further away from the traffic jam, the by-passroute length increase could be reduced significantly. The breakpoints in the graph are thepoints in the road network, where a new by-pass route could be taken. As we can see, theoptimal query distance in this case is around 1000 meters. That is the point where the vehiclescan take the shortest by-pass route according to the information contained in the returned TTImessage.On the diagrams noted with (M2), a major road in the city is displayed, where the length ofthe original route was 2500 meters. It can be seen that in case of short query distances thelength of the by-pass routes could be as much as the double of the original route length. Asthe query distance is increased to 1000 meters, the by-pass route length decreases to around600 meters, and with a query distance of 1200 meters, the route length increase is only 100meters. This shows that finding the optimal query distance is really important, because thelength of the by-pass route can be reduced significantly.The diagrams notedwith (M1) present a case when the traffic jam is on amain downtown roadwith plenty of nearby roads, which can be taken as by-pass routes. Thus, the query distancecan be set to a small value, since the increase in by-pass route will become negligible.

4.4 Comparison of SPACE and SPACE ILPUntil now we investigated the effect of traffic congestion on TTI dissemination separatelystudying the heuristic (SPACE) and the optimal (SPACE ILP) Domain of Interest calculationalgorithms. Considering the results from Section 4.1 and Section 4.2 we can affirm that theoutcome of the algorithms present certain similarities. For example, from both approachesit turns out that the traffic jams can be classified in two major categories. One categoryis represented by traffic jams of main, crucial roads (e.g. bridge), with a large Domain ofInterest and an increased length of the by-pass routes. The dissemination of TTI for suchtraffic incidents is extremely important, since the zone of no return of these traffic jams is alsolarge. The second category of traffic jams is represented by downtown roads with small DoI


www.intechopen.com


Fig. 6. Query results of different traffic jams

values. Such congestions can be avoided quite easily, since there is a large number of shorterby-pass routes around them.Besides discovering resemblances of the algorithms’ outcome it is also important to considertheir potential benefits and differences. The main advantage of SPACE ILP is the optimal sizeof DoI result sets for different parameters of the actual context (e.g. network load, reflected byparameter α). However, the SPACE ILP algorithm can be employed only for the calculationof a limited number of DoI calculations, since its running time is relatively high; thus, itcannot be used to calculate the DoI for all the segments of a larger road network. Thatis where the SPACE algorithm comes into picture, because it presents much faster runningtimes. Unfortunately, there is no method defined on how to select the most important roadsegments from the outcome of the SPACE algorithm. The impact factor metric gives us a goodmeasure regarding the importance of different road segments, but it does not indicate a certainthreshold , which would limit the DoI area for the respective result set. Therefore, the aim ofthis subsection is to provide a method, which provides a relationship, associates the results ofthe two DoI generator algorithms.In order to find such a relationship we opted to compare and analyze the result sets of thetwo algorithms in a spatial manner. We designed and built an extension for our RUBeNSvehicular simulation environment Laborczi et al. (2006), which is able to load, store andanalyze the DoI result sets of the algorithms. The extension is built in the PostgreSQLdatabasemanagement system and takes advantage of its geographic support, PostGIS and pgrouting.In this framework by using common map references we have an unified view of uploadeddata, and through embedded functions we are able to define different spatial operations andmethods for analyzing the uploaded result sets. For example, we can calculate the area orperimeter of DoIs, we can compare the DoI result sets regarding their spatial coverage andrelationships, and we can design methods to intelligently reduce the area of SPACE DoIs.For the reduction of SPACE DoI areas we implemented the following simple method. Fromthe uploaded DoI sets of SPACE ILP we selected certain cases, which represent characteristicDoIs (for large and small traffic jams). From these DoIs we get the optimized set of roadsegments, which in turn will be searched in the DoI set of the SPACE algorithm for the sametraffic jam. Based on this we get a reduced set of road segments from the respective SPACE


www.intechopen.com


Fig. 7. Comparison of Domains of Interest of the SPACE and SPACE ILP algorithms

DoI, for which the dissemination area is optimal. Then considering the impact factors anddomain of interest lengths for this result set we calculate the derivative of the graph’s slope.This will provide the threshold for limiting the area of other DoI result sets. For reducingthe size of other SPACE DoI result sets we always select road segments until their graph’sslope represents higher values than the calculated threshold. This would mean that theseroad segments are still important for TTI dissemination about the respective traffic incident.By using this calculation the size of all SPACE DoIs can be reduced; thus, a correspondingassociation between the two algorithms was identified.The results of the method are presented on Figure 7, where we represent how the roadsegments with different relevance regarding the DoI are situated along a selected vehicle’sroute. A cross-bridge route was selected, where the length of DoI (considered only on thisspecific path) is represented in function of the link IDs along the path, the traffic jams weregenerated consecutively for the respective road segments. On the figure traffic jams with largeinfluence (large DoI) can be observed between links with IDs 113870 and 114205 (critical part),where the DoI (along the route) of SPACE can reach even 1400 meters. This means that in caseof a traffic jam situated along this critical part the TTI should be disseminated to a large partof the route, in order to avoid the traffic jam of the respective links with small by-pass routes.This critical part of the route contains also the bridge. For the rest of the route the congestedlinks can be avoided easily, this is represented by small values for DoI.It is important to observe that the results of DoI sets for SPACE and SPACE ILP representsimilar behavior, emphasizing the difference between the different kinds of traffic jams (withsmall, respectively large DoI values). The difference between the values of DoI length of thealgorithm’s output come from the fact that in the case of the SPACE algorithm we addeda few more link segments (additional length increase), since we wanted to provide a largerzone for query and decision making during the trip along the respective route segment. Theoutcome of SPACE ILP is a little bit to optimistic, since it does not take into account the delaysof information propagation.By applying the method for the whole set of uploaded SPACE DoIs (Figure 8) we can observethat size of DoIs (original, without reduction) shifts from the larger values towards muchsmaller ones (reduced DoIs). This can be attributed to the fact that only a few road segments


www.intechopen.com


Fig. 8. Reshaping the Domains of Interest of the SPACE algorithm

are really important (large DoI areas), while the majority of traffic jams affect roads withsmall influence (from downtown areas). Similar results could be achieved by employing theSPACE ILP algorithm for DoI calculations. Thus, the SPACE DoI reduction algorithm provesto be effective to determine the correct size of dissemination areas.

5. Conclusions

The current trends and problems of intelligent transportation systems have been presentedin the beginning of this chapter. We presented a short overview of research on providinglocation-aware services in vehicular networks by disseminating information messages, forexample regarding traffic congestion or fuel prices at a gas station, and maintaining thesemessages at key areas in the traffic network. This way the vehicles that traverse throughthese areas get informed about the content of the message and are able to alter their routeaccordingly. In the second part of the paper we presented SPACE, a heuristic algorithmto determine the domains of interest to a given event, for example a traffic jam, wherethe dissemination of the message is important. Following that we have given a linearprogramming formulation to determine the optimal domains of interest for a given trafficscenario. In the next section an extension of the linear programming formulation wasdescribed, to take the velocity of the vehicles and the message propagation delay intoaccount. This allowed the extension of the DoI, so the vehicles could be notified beforethey reach the junction where the optimal alternative route starts. In the final section weevaluated the heuristic and the linear programming algorithms with different settings of theadjustable parameters using the RUBeNS simulation environment. It was shown, that thelinear programming solution can be used to calibrate the heuristic algorithm, although SPACEdoes not presents the optimal solution like the linear ILP algorithm, but it runs significantlyfaster, thus it is more usable.

6. References

Burgess, J., et al.,MaxProp: Routing for Vehicle-Based Disruption-TolerantNetworks, IEEE Infocom2006, 2006 April, Barcelona, Spain.


www.intechopen.com


Dikaiakos, M. D., et al. (2007), Location-Aware Services over Vehicular Ad-Hoc Networks usingCar-to-Car Communication, IEEE Journal on Selected Areas in Communications, vol.25, no. 8, October 2007.

Dikaiakos, M. D., et al. (2010), On the Evaluation of Caching in Vehicular Information Systems, 9thHellenic Data Management Symposium; 2010 July 2.

Laborczi, P., et al., In: Vehicle-to-Vehicle Traffic Information System with Cooperative RouteGuidance, 13th World Congress on Intelligent Transport Systems; 2006 October 8-12;London, United Kingdom.

Laborczi, P., Mezny B., Torok, A., Ruzsa, Z., Query-based Information Gathering in IntelligentTransportation Systems, International Symposium on Combinatorial Optimization,March 24-26, 2010, Hammamet, Tunisia

Leontiadis, I., Mascolo, C., Opportunistic SpatioTemporal Dissemination System for VehicularNetworks, The First International Workshop on Mobile Opportunistic NetworkingACM/SIGMOBILE MobiOpp 2007; 2007 June 11; San Juan, Puerto Rico, USA.

Lochert C., et al., Data Aggregation and Roadside Unit Placement for a VANET Traffic InformationSystem, ACM VANET’08, September 15, 2008, San Francisco, California, USA.

Sormani, D., et al., Towards Lightweight Information Dissemination in Inter-Vehicular Networks,The Third ACM International Workshop on Vehicular Ad Hoc Networks; 2006September 29; Los Angeles, California, USA.

Speieys, L., Jensen, C. S., Enabling Location-based Services Multi-Graph Representation ofTransportation Networks, Journal GeoInformatica, Publisher Springer Netherlands, ISSN1384-6175

Tian, J., Han, L., Rothermel, K., Spatially Aware Packet Routing for Mobile Ad Hoc Inter-VehicleRadio Networks, IEEE Intelligent Transportation Systems, 2004, 2004 October 12-15.

Torok, A., Laborczi, P., Gerhath, G., Spatially Constrained Dissemination of Traffic Informationin Vehicular Ad Hoc Networks, IEEE Vehicular Technology Conference VTC2008-Fall,21-24 Sept., 2008, Calgary, Canada.

Torok, A., Laborczi, P., Mezny, B., Context-aware Traffic Information Flooding in Vehicular Ad HocNetworks, Tamkang Journal of Science and Engineering, Vol. 13, No. 1, 2010

Zhao, J., Cao, G., VADD: Vehicle-Assisted Data Delivery, Vehicular Ad Hoc Networks, IEEEInfocom 2006, 2006 April, Barcelona, Spain.


www.intechopen.com

Mobile Ad-Hoc Networks: ApplicationsEdited by Prof. Xin Wang

ISBN 978-953-307-416-0Hard cover, 514 pagesPublisher InTechPublished online 30, January, 2011Published in print edition January, 2011

InTech EuropeUniversity Campus STeP Ri Slavka Krautzeka 83/A 51000 Rijeka, Croatia Phone: +385 (51) 770 447 Fax: +385 (51) 686 166www.intechopen.com

InTech ChinaUnit 405, Office Block, Hotel Equatorial Shanghai No.65, Yan An Road (West), Shanghai, 200040, China

Phone: +86-21-62489820 Fax: +86-21-62489821

Being infrastructure-less and without central administration control, wireless ad-hoc networking is playing amore and more important role in extending the coverage of traditional wireless infrastructure (cellularnetworks, wireless LAN, etc). This book includes state-of the-art techniques and solutions for wireless ad-hocnetworks. It focuses on the following topics in ad-hoc networks: vehicular ad-hoc networks, security andcaching, TCP in ad-hoc networks and emerging applications. It is targeted to provide network engineers andresearchers with design guidelines for large scale wireless ad hoc networks.

How to referenceIn order to correctly reference this scholarly work, feel free to copy and paste the following:

Attila Török, Balázs Mezny and Peter Laborczi (2011). Traffic Information Dissemination in Vehicular Ad HocNetworks, Mobile Ad-Hoc Networks: Applications, Prof. Xin Wang (Ed.), ISBN: 978-953-307-416-0, InTech,Available from: http://www.intechopen.com/books/mobile-ad-hoc-networks-applications/traffic-information-dissemination-in-vehicular-ad-hoc-networks

© 2011 The Author(s). Licensee IntechOpen. This chapter is distributedunder the terms of the Creative Commons Attribution-NonCommercial-ShareAlike-3.0 License, which permits use, distribution and reproduction fornon-commercial purposes, provided the original is properly cited andderivative works building on this content are distributed under the samelicense.

https://creativecommons.org/licenses/by-nc-sa/3.0/

Documents

y Traf Þ c Information Dissemination in Vehicular Ad Hoc ... · Traf c and Travel Information (TTI) spreading in vehicular ad hoc networks is achieved by the means of a oodingmechanism